• =?UTF-8?B?Ik5vcmRsw7Z3Ig==?= (7/7) Mar 30 2014 Does DMD currently avoid range checks in array slice expressions
• bearophile (6/10) Mar 30 2014 You have two simple ways to answer your question: try to go past
• Steven Schveighoffer (11/16) Mar 31 2014 as
• =?UTF-8?B?Ik5vcmRsw7Z3Ig==?= (6/8) Mar 31 2014 $/n is of course always within range if n is positive integer >=
• =?UTF-8?B?Ik5vcmRsw7Z3Ig==?= (2/3) Mar 31 2014 That would be an interesting task to fix :)

=?UTF-8?B?Ik5vcmRsw7Z3Ig==?= <per.nordlow gmail.com> writes:

Does DMD currently avoid range checks in array slice expressions 
such as

f(x[0..$/2])
f(x[$/2..$])

typically found in divide-and-conquer algorithms such as 
quicksort?

If not, what would it require to implement it?

"bearophile" <bearophileHUGS lycos.com> writes:

Nordlöw:

 Does DMD currently avoid range checks in array slice 
 expressions such as

 f(x[0..$/2])
 f(x[$/2..$])

You have two simple ways to answer your question: try to go past 
the slices and look if D raises an error, and look at the asm 
produced with various compiler switches.

Bye,
bearophile

"Steven Schveighoffer" <schveiguy yahoo.com> writes:

On Sun, 30 Mar 2014 15:40:43 -0400, Nordl=C3=B6w <per.nordlow gmail.com>=
 wrote:

 Does DMD currently avoid range checks in array slice expressions such =

as
 f(x[0..$/2])
 f(x[$/2..$])

 typically found in divide-and-conquer algorithms such as quicksort?

If they are range-checked, it would be a good addition to the optimizer =
to  =

remove them. My guess is that it is range-checked.

 If not, what would it require to implement it?

As a hack, you can use the pointer, which will not be range checked:

f(x.ptr[0..x.length/2]) // note you can't use $ because the pointer  =

doesn't have length
f(x.ptr[x.length/2..x.length])

-Steve

=?UTF-8?B?Ik5vcmRsw7Z3Ig==?= <per.nordlow gmail.com> writes:

 If they are range-checked, it would be a good addition to the 
 optimizer to remove them. My guess is that it is range-checked.

$/n is of course always within range if n is positive integer >= 
1.

But what about in the general indexing/slicing case? In that case 
it would be useful if we could reuse value range propagation 
(VRP) in DMD, to figure out which other expressions that don't 
need range checking.

=?UTF-8?B?Ik5vcmRsw7Z3Ig==?= <per.nordlow gmail.com> writes:

 If not, what would it require to implement it?


That would be an interesting task to fix :)

DMD source references anyone?